The coordination of probabilistic inference in neural systems

Abstract

Life, thought of as adaptively organised complexity, depends upon information and inference, which is nearly always inductive, because the world, though lawful, is far from being wholly predictable. There are several influential theories of probabilistic inference in neural systems, but here I focus on the theory of Coherent Infomax, and its relation to the theory of free energy reduction. Coherent Infomax shows, in principle, how life can be preserved and improved by coordinating many concurrent inferences. It argues that neural systems combine local reliability with flexible, holistic, context-sensitivity. What this perspective contributes to our understanding of neuronal inference is briefly outlined by relating it to cognitive and neurophysiological studies of context-sensitivity and gain-control, psychotic disorganization, theories of the Bayesian brain, and predictive coding. Limitations of the theory and unresolved issues are noted, emphasizing those that may be of interest to philosophers, and including the possibility of major transitions in the evolution of inferential capabilities.